Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method of image labeling comprising: receiving at least one image; computing, by a first stage predictor, at least one global parameter of the image; and modifying an image labeling process according to the computed global parameter, the at least one global parameter being one of: a game context action, age, and gender, modifying the image labeling process including dynamically modifying a second stage predictor using the computed global parameter; wherein computing the global parameter comprises: selecting a subset of image elements in the image; for each selected image element determining a probability that the image element depicts an object having a specified global parameter; and combining the probability from a plurality of the selected image elements to compute an estimate of the global parameter, wherein computing the global parameter also includes ensuring all available global parameter values are associated with a non-zero probability.
Image labeling is performed by first receiving an image and then computing a "global parameter" (e.g., game context action, age, gender) using a first stage predictor. This global parameter is used to modify the image labeling process, which involves dynamically modifying a second stage predictor. Calculating the global parameter involves selecting a subset of image elements, determining the probability each element represents an object with a specific global parameter value, and combining these probabilities to estimate the global parameter. The process ensures all possible global parameter values have a non-zero probability.
2. A method as claimed in claim 1 wherein modifying the image labeling process comprises rectifying the at least one input image using the at least one computed global parameter.
The image labeling process described in Claim 1 modifies the image labeling process by adjusting the input image based on the calculated global parameter. This involves rectifying or transforming the input image, ensuring consistency in scale or orientation, using the computed global parameter.
3. A method as claimed in claim 1 wherein modifying the image labeling process comprises dynamically modifying at least one stored parameter of the image labeling process during use of the image labeling process to label the image.
The image labeling process described in Claim 1 modifies the image labeling process by dynamically changing at least one stored parameter of the labeling process while the image is being labeled. This enables adaptive modification during the real-time labeling process, using the computed global parameter.
4. A method as claimed in claim 1 wherein the image labeling process comprises a random forest and modifying the image labeling process comprises using different distributions stored at leaf nodes of the random forest or changing feature tests and thresholds at nodes of the random forest.
The image labeling process described in Claim 1 uses a random forest algorithm. Modifying the image labeling process involves using different probability distributions at the leaf nodes of the random forest or changing feature tests and threshold values at the nodes within the forest based on the global parameter.
5. A method according to claim 1 wherein computing the global parameter comprises using a classification forest comprising a plurality of classification trees.
The image labeling process described in Claim 1 computes the global parameter using a classification forest which is made up of multiple classification trees.
6. A method according to claim 1 wherein computing the global parameter comprises using a regression forest comprising a plurality of regression trees.
The image labeling process described in Claim 1 computes the global parameter using a regression forest which is made up of multiple regression trees.
7. A method according to claim 5 wherein combining the probabilities from a plurality of the selected image elements to compute an estimate of the global parameter comprises aggregating the probabilities stored at all of the leaf nodes in the forest for a set of image elements.
The image labeling process using classification trees, as described in Claim 5, computes the global parameter by combining probabilities. It aggregates probabilities stored at all leaf nodes within the classification forest for a set of selected image elements.
8. A method according to claim 6 wherein combining the probabilities from a plurality of the selected image elements to compute an estimate of the global parameter comprises using the stored probabilities from a subset of leaf nodes wherein the subset of leaf nodes comprises the most confident leaf nodes within the regression forest.
The image labeling process using regression trees, as described in Claim 6, computes the global parameter by combining probabilities. It uses stored probabilities from a subset of leaf nodes, focusing on the most confident leaf nodes within the regression forest.
9. A method according to claim 1 wherein the image labeling process is trained using a set of training data comprising images with specified labels and where the set of training data comprises only images which have had the variations of at least one global parameter accounted for, either by image rectification or by feature adaptation.
The image labeling process described in Claim 1 is trained with training data containing images with specified labels. The training data consists of images where variations due to at least one global parameter have already been addressed either by image rectification or feature adaptation.
10. A method as claimed in claim 1 wherein the image labeling process is trained using a set of training data comprising images with specified labels and wherein the method comprises rectifying the training data using the at least one global parameter.
The image labeling process described in Claim 1 is trained with images and specified labels. The training images are rectified using the calculated global parameter before training the image labeling process.
11. A method of pose estimation comprising; receiving at least one image depicting at least part of one human or at least part of one animal; computing, by a first stage predictor, for the at least one image at least one global parameter describing image level characteristics of the at least one human or animal, the at least one global parameter being age; and rectifying the at least one image using the global parameter prior to inputting that image to an image labeling process to obtain an estimate of a pose of the human or animal, the image labeling process including using a second stage predictor that is trained using training data that is rectified using the global parameter; wherein computing the global parameter comprises; selecting a subset of image elements in the image; for each selected image element determining a probability that the image element depicts an object having a specified global parameter; and combining the probabilities from a plurality of the selected image elements to compute an estimate of the global parameter, wherein computing the global parameter also includes ensuring all available global parameter values are associated with a non-zero probability.
Pose estimation involves receiving an image of a human or animal and computing a global parameter (specifically, age) using a first stage predictor. The image is then rectified using the computed age parameter before being processed by an image labeling algorithm (second stage predictor) to estimate the pose. The second stage predictor is trained with training data that has also been rectified using the age parameter. Computing the age parameter involves selecting a subset of image elements, determining the probability each element depicts an object with a certain age, and combining these probabilities to estimate the age. The process ensures all possible age values have a non-zero probability.
12. A method as claimed in claim 11 wherein computing the global parameter comprises using at least one of; a classifier, support vector machine, regression forest, classification forest, relevance vector machine; having been trained to predict a global parameter, using a set of labeled training data.
In the pose estimation method described in Claim 11, computing the global parameter involves using a classifier, support vector machine, regression forest, classification forest, or relevance vector machine. These models are pre-trained to predict the global parameter using labeled training data.
13. A method as claimed in claim 11 further comprising computing a second global parameter; wherein computing the second global parameter comprises using at least one of a regression forest or a classification forest which comprises a plurality of regression or classification trees trained to predict global parameters, wherein global parameter predictions for each of the global parameters are stored at each leaf node of each tree.
The pose estimation method described in Claim 11 also computes a second global parameter. Computing this second global parameter utilizes either a regression forest or a classification forest, each comprising trees trained to predict global parameters. The global parameter predictions for both global parameters are stored at the leaf nodes of the trees.
14. A method as claimed in claim 11 wherein computing a global parameter comprises using at least one of; a classifier, support vector machine, regression forest, classification forest, relevance vector machine; having been trained using one of recursive, iterative or breadth first training.
In the pose estimation method described in Claim 11, computing the global parameter utilizes a classifier, support vector machine, regression forest, classification forest, or relevance vector machine that has been trained using recursive, iterative, or breadth-first training methods.
15. A method as claimed in claim 13 wherein the regression or classification forest is trained using a first set of training data for the first global parameter being predicted; and the training is repeated using a second set of training data for the second global parameter being predicted.
In the pose estimation method described in Claim 13, the regression or classification forest is trained using a first set of training data to predict the first global parameter. The training process is then repeated with a second, possibly different, set of training data to predict the second global parameter.
16. A human or animal pose estimation system comprising: at an input arranged to receive at least one image depicting at least part of a human or at least part of an animal; and a processor arranged to compute at least one global parameter of the image using a first stage predictor, the at least one global parameter being a game context action, the processor being arranged to: select a subset of image elements in the image; for each selected image element determine a probability that the image element depicts an object having a specified global parameter; and combine the probability from a plurality of the selected image elements to compute an estimate of the global parameter, the first stage predictor ensuring all available global parameter values are associated with a non-zero probability; the processor being further arranged to modify an image labeling process of a second stage predictor according to the computed global parameter by using different distributions stored at leaf nodes of a random forest based on the computed global parameter or by changing, based on the computed global parameter, features tests and thresholds at nodes of the random forest to obtain an estimate of a pose of the human or animal depicted in the image.
A pose estimation system for humans or animals receives an image and a processor computes a global parameter (specifically, a game context action) using a first stage predictor. The processor selects image elements, determines the probability each element depicts an object related to a specific action, and combines these probabilities to estimate the global parameter. The first stage predictor ensures all possible game context actions have a non-zero probability. The processor modifies an image labeling process of a second stage predictor based on the computed action by using different distributions at leaf nodes of a random forest, or by changing feature tests and thresholds at nodes within the random forest, to estimate the pose.
17. A system according to claim 16 wherein the at least one image is received from one of; a depth camera and a medical imaging device.
The pose estimation system described in Claim 16 receives the image from either a depth camera or a medical imaging device.
18. A gaming apparatus comprising a human or animal pose estimation system as claimed in claim 16 .
A gaming apparatus includes the human or animal pose estimation system as described in Claim 16.
19. A gaming apparatus as claimed in claim 16 which is arranged to influence the course of a game using the estimated pose of the human or animal.
The gaming apparatus described in Claim 16 utilizes the estimated pose of the human or animal to influence the game's progression.
Unknown
October 14, 2014
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.